AI Model Outperforms Human Doctors in Emergency Room Diagnosis Study at Harvard
Harvard study found that at least one large language model demonstrated greater diagnostic accuracy than human doctors in emergency room cases.
Sources · TechCrunch AI
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
861–880 of 1,902Last Week in AI Podcast Discusses GPT 5.5, DeepSeek V4, and AI Safety Concerns
The 243rd episode of the Last Week in AI podcast covers recent developments including GPT 5.5, DeepSeek V4, and AI safety issues.
Sources · Last Week in AI
ML-Agent: Researchers Introduce Reinforcement Learning Framework for Autonomous Machine Learning Engineering
New framework uses online RL to train 7B LLM agents for autonomous ML tasks, achieving performance comparable to larger proprietary models.
Sources · arxiv.org
New Memory-Based Frameworks Enhance LLM Agent Learning and Reasoning
Researchers propose memory architectures using episodic and semantic storage to improve LLM adaptation without parameter updates.
Sources · arxiv.org
Researchers Propose New Methods to Improve Language Model Training Through Reasoning-Aware Optimization
Two new arXiv papers present techniques for enhancing LLM training by incorporating reasoning structure and step-level importance into optimization.
Sources · arxiv.org
Disneyland Implements Face Recognition Technology for Visitors
Disneyland has begun using facial recognition technology on park visitors, according to reporting from WIRED.
Sources · WIRED AI
Four New Multi-Agent AI Frameworks Target Optimization Challenges Across Diverse Domains
Researchers published four separate multi-agent AI systems on arXiv, each addressing complex optimization tasks in recommendation systems, materials science, and biology.
Sources · arxiv.org
Four New Papers Tackle Reasoning, Bias, and Learning Challenges in Large Language Models
Researchers present new approaches for mitigating selection bias, separating reasoning modes, context learning, and self-supervised problem generation.
Sources · arxiv.org
Meta Acquires Robotics Startup Assured Robot Intelligence
Meta has purchased humanoid robotics startup Assured Robot Intelligence to enhance its AI models for robots, according to the company.
Sources · TechCrunch AI
Musk Testifies in OpenAI Lawsuit, Claims He Was Deceived and Admits xAI Uses OpenAI Models
Elon Musk testified in his lawsuit against OpenAI, claiming he was duped into funding the company and revealing xAI trained on OpenAI's models.
Sources · nytimes.com / technologyreview.com +1
New Research Explores LLM Agents for Engineering Optimization and Scientific Tasks
Four new papers examine AI agents for chip design, tool-calling evaluation, scientific visualization, and engineering methodology.
Sources · arxiv.org
Amazon AWS Explores Reinforcement Learning with LLM-as-a-Judge for Nova Models
AWS published details on using reinforcement learning with AI-as-judge feedback for fine-tuning Amazon Nova language models.
Sources · Amazon AWS AI
Goodfire Releases Silico Tool for Debugging Large Language Models
San Francisco startup Goodfire has launched Silico, a mechanistic interpretability tool that allows researchers to examine and adjust AI model parameters during training.
Sources · MIT Technology Review
Musk Testifies xAI Used OpenAI Models for Training; Zilis Emerges as Key Figure in Trial
Elon Musk acknowledged in court that xAI partly used OpenAI's models to train its own, while testimony reveals Shivon Zilis's role as liaison.
Sources · wired.com / techcrunch.com
Researchers Develop Multi-Agent AI Frameworks for Recommendation Systems, Material Discovery, and Biological Prediction
Four new research papers introduce multi-agent AI systems leveraging large language models for complex optimization and prediction tasks.
Sources · arxiv.org
Researchers Introduce New LLM Agent Systems for Chip Design, Tool-Calling, and Scientific Workflows
Four new research papers explore LLM agent capabilities in chip optimization, inference-time feedback, scientific visualization, and engineering methodology.
Sources · arxiv.org
Anthropic Could Raise $50B at $900B Valuation, Sources Say
The Claude AI maker has received multiple pre-emptive funding offers at valuations between $850B and $900B, according to sources.
Sources · TechCrunch AI
ChatGPT Images 2.0 Shows Improved Text Generation, Alibaba Releases Qwen 3.6 Max Preview
ChatGPT's Images 2.0 demonstrates enhanced text generation capabilities while Alibaba launches Qwen 3.6 Max Preview model.
Sources · Last Week in AI
Elon Musk Testifies About 2017 OpenAI Power Struggle and Funding Halt
Musk faced cross-examination about emails showing he demanded board control and stopped funding OpenAI in 2017.
Sources · wired.com / techcrunch.com
New Multi-Agent AI Systems Show Promise in Cybersecurity, Code Generation, and Coordination
Researchers present advances in multi-agent AI across cybersecurity troubleshooting, test-driven development, and emergent coordination capabilities.
Sources · arxiv.org